Your local GPU can't run a 400B parameter model, but its compute can pay for one. Hashference lets developers and AI agents pay for 45+ frontier models (Claude, GPT, DeepSeek) using background CPU/GPU work, Stratum mining, or HTTP 402 PoW.